Anti-interference serial communication method and system of industrial-grade wireless module

Abstract

The application provides an anti-interference serial communication method and system of an industrial-grade wireless module, comprising: a multi-dimensional interference feature acquisition module, which acquires interference data in serial communication of the industrial-grade wireless module in real time to form a multi-scale interference feature vector; the multi-scale interference feature vector is input into a Mamba communication state space model with a fusion CRC check mechanism, a potential interference mode is predicted through time sequence modeling of linear complexity; and a transmission parameter under the potential interference mode is dynamically weighted and adjusted based on an adaptive weight transmission optimization function, wherein the transmission parameter comprises a baud rate, a data bit length, a check mode and a retransmission strategy. In the application, the cost of the anti-interference serial communication is reduced without increasing the hardware cost; meanwhile, the transmission parameter under the potential interference mode is dynamically weighted and adjusted based on the adaptive weight transmission optimization function, so that real-time tracking of dynamic changes of interference is realized to realize anti-interference communication.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication devices, in particular to an anti-interference serial port communication method and system of an industrial-grade wireless module. BACKGROUND

[0002] Serial port communication, as one of the most basic communication methods in industrial Internet of Things, is widely used in sensor data acquisition, device control, remote monitoring and other scenarios. Industrial-grade wireless modules are increasingly widely used in intelligent manufacturing, smart grid, industrial automation and other fields, and the electromagnetic environment they face is becoming increasingly complex. In industrial sites, the start and stop of motors, the operation of frequency converters, and the operation of radio frequency devices can all produce strong electromagnetic interference, seriously affecting the stability and reliability of serial port communication.

[0003] The anti-interference method of serial port communication of traditional industrial-grade wireless modules mainly focuses on hardware protection and fixed parameter configuration. At the hardware level, measures such as shielding cables, adding filter circuits, and using isolation chips are usually used to reduce the impact of external electromagnetic interference on communication signals. For example, in RS-485 communication, TVS diodes are connected in parallel at both ends of the transceiver for transient overvoltage protection to suppress surge interference. At the software level, fixed check methods (such as parity check) and retransmission strategies are often used, and a limited number of retransmissions are performed when errors are detected.

[0004] In the above-mentioned anti-interference method of serial port communication, the use of hardware measures such as multi-layer shielding cables and high-isolation chips significantly increases the cost of the module, which is not conducive to large-scale application. Once the hardware design is determined, it is difficult to flexibly respond to different types and intensities of interference, especially new or sudden interference sources. Additional hardware circuits increase the size of the module and also increase power consumption, which is not in line with the development trend of small size and low power consumption of industrial-grade wireless modules.

[0005] As for the software level, an anti-interference serial port parameter adjustment mechanism based on interference levels is proposed. This type of method usually presets several fixed parameter configurations, and the fixed strategy usually switches parameters based on preset thresholds, which cannot track the dynamic changes of interference in real time and is not effective in scenarios with rapid fluctuations in interference. SUMMARY

[0006] The main purpose of the present application is to provide an anti-interference serial port communication method and system of an industrial-grade wireless module, aiming to reduce the cost of anti-interference serial port communication and track the dynamic changes of interference in real time for anti-interference communication.

[0007] To achieve the above-mentioned purpose, the present application provides an anti-interference serial port communication method of an industrial-grade wireless module, comprising the following steps:

[0008] Based on the multi-dimensional interference feature acquisition module, the interference data in the industrial-grade wireless module serial communication is collected in real time to form a multi-scale interference feature vector.

[0009] The multi-scale interference feature vector is input into the Mamba communication state space model with a fusion CRC check mechanism, and the potential interference mode is predicted through linear complexity time sequence modeling.

[0010] Based on the adaptive weight transmission optimization function, the transmission parameters under the potential interference mode are dynamically weighted and adjusted; wherein the transmission parameters include baud rate, data bit length, check mode and retransmission strategy.

[0011] Further, the interference data includes interference signal strength, bit error rate, clock offset and electromagnetic radiation spectrum data.

[0012] Further, the multi-dimensional interference feature acquisition module adopts a parallel multi-resolution feature processing structure, which includes:

[0013] High-frequency interference feature branch, for capturing transient electromagnetic interference signals above 100kHz;

[0014] Medium-frequency communication quality branch, for analyzing the bit error rate fluctuation law of 10kHz-100kHz;

[0015] Low-frequency clock offset branch, for monitoring the clock synchronization deviation below 10kHz;

[0016] Each branch realizes information fusion through a cross-scale feature interaction module to obtain a multi-scale interference feature vector.

[0017] Further, the Mamba communication state space model is specifically used for:

[0018] A deep separable convolution module is used to extract local features from the input multi-scale interference feature vector to obtain a reduced dimension interference feature;

[0019] Through the dynamic weight calculation unit of the Mamba state space module, the long-distance dependence modeling is performed on the preset historical interference sequence and the reduced dimension interference feature, and the linear complexity time convolution operation is used to predict the interference mode in the next 5 to 10 communication periods;

[0020] The check result of the traditional CRC check code is input as a residual error, which is bit-by-bit fused with the predicted interference mode, and the prediction deviation is corrected through the residual connection mechanism to obtain the potential interference mode; wherein the generating polynomial of the CRC check is based on the preset industrial standard.

[0021] Further, the Mamba communication state space model is further enhanced by RMSNorm normalization layer and SiLU activation function to express nonlinearity, and the overall parameter quantity is controlled within 1MB, which is suitable for the embedded computing resources of industrial wireless modules.

[0022] Further, the adaptive weight transmission optimization function expression is:

[0023]

[0024] wherein, is the bit error rate loss term, which uses Smooth L1 loss to measure the deviation of the actual bit error rate from the threshold value;

[0025] is the serial port resource consumption loss term, which is used to constrain the power consumption fluctuation caused by the baud rate adjustment;

[0026] is the protocol compatibility loss term, which ensures that the parameter adjustment conforms to the industrial communication protocol specification;

[0027] is the dynamic weight coefficient, which automatically adjusts the proportion according to the real-time interference intensity.

[0028] Further, the adjustment strategy of the dynamic weight coefficient includes:

[0029] When the electromagnetic interference intensity exceeds 80dB, set ;

[0030] When the electromagnetic interference intensity is less than 40dB, set ;

[0031] When the electromagnetic interference intensity is not less than 40dB and not more than 80dB, set

[0032] wherein, the adjustment period of the weight coefficient is inversely proportional to the baud rate of the serial communication.

[0033] Further, based on the adaptive weight transmission optimization function, the transmission parameters under potential interference mode are dynamically weighted and adjusted, including:

[0034] Obtain the electromagnetic interference intensity, based on the mapping rule of interference intensity and weight coefficient, set the proportion of reliability weight α, efficiency weight β, and compatibility weight γ; wherein α corresponds to the bit error rate constraint intensity, β corresponds to the adjustment priority of baud rate and data bit length, and γ corresponds to the industrial protocol compatibility verification weight;

[0035] Determine the differentiated adjustment strategy for different transmission parameters:

[0036] Baud rate adjustment: when the alpha weight ratio is greater than or equal to 3, a stepwise speed reduction mechanism is started, and the current baud rate is adjusted by 90%, 80%, and 70% grading, and the interval between each grading is not more than 20ms;

[0037] Data bit length configuration: when beta is greater than 4, 8-bit data bits are used to improve transmission efficiency, otherwise 7-bit data bits are used to reduce error code risk;

[0038] Check mode switching: when gamma is greater than or equal to 3, CRC-16 check is forced to be enabled and two parity check bits are added;

[0039] Re-transmission strategy optimization: based on the weight difference between alpha and beta, the retransmission interval is dynamically adjusted, when the difference is greater than 1, the retransmission interval is shortened from 50ms to 30ms, and when the difference is less than or equal to 1, the default interval is maintained;

[0040] The real-time effect of parameter adjustment is realized through a hardware register mapping table, and the weight calculation result is converted into a configuration instruction of a serial port controller.

[0041] The application also provides an anti-interference serial communication system of an industrial-grade wireless module, comprising:

[0042] The acquisition unit is used for acquiring interference data in industrial-grade wireless module serial communication in real time based on a multi-dimensional interference feature acquisition module, and forming a multi-scale interference feature vector;

[0043] The prediction unit is used for inputting the multi-scale interference feature vector into a Mamba communication state space model of a fusion CRC check mechanism, and predicting a potential interference mode through linear complexity time sequence modeling;

[0044] The adjustment unit is used for dynamically weighting and adjusting the transmission parameters under the potential interference mode based on an adaptive weight transmission optimization function; wherein the transmission parameters include baud rate, data bit length, check mode and retransmission strategy.

[0045] The application also provides a computer device comprising a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the steps of the method of any one of the above.

[0046] The application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the steps of the method of any one of the above.

[0047] This invention provides an anti-interference serial communication method and system for industrial-grade wireless modules, comprising: real-time acquisition of interference data in serial communication of industrial-grade wireless modules based on a multi-dimensional interference feature acquisition module, forming a multi-scale interference feature vector; inputting the multi-scale interference feature vector into a Mamba communication state space model incorporating a CRC check mechanism, and predicting potential interference modes through linear complexity time-series modeling; and dynamically adjusting transmission parameters under potential interference modes based on an adaptive weighted transmission optimization function; wherein the transmission parameters include baud rate, data bit length, check method, and retransmission strategy. In this invention, no additional hardware cost is required, reducing the cost of anti-interference serial communication; simultaneously, based on the adaptive weighted transmission optimization function, the dynamic weighted adjustment of transmission parameters under potential interference modes enables real-time tracking of dynamic changes in interference for anti-interference communication. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the anti-interference serial communication method steps of an industrial-grade wireless module in one embodiment of the present invention;

[0049] Figure 2 This is a block diagram of an anti-interference serial communication system for an industrial-grade wireless module according to an embodiment of the present invention.

[0050] Figure 3 This is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0051] The implementation, functional features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0053] Reference Figure 1 One embodiment of the present invention provides an anti-interference serial communication method for an industrial-grade wireless module, comprising the following steps:

[0054] Step S1: Based on the multi-dimensional interference feature acquisition module, interference data in the serial communication of industrial-grade wireless modules is collected in real time to form a multi-scale interference feature vector.

[0055] Step S2: Input the multi-scale interference feature vector into the Mamba communication state space model that integrates CRC check mechanism, and predict potential interference patterns through time-series modeling with linear complexity.

[0056] In step S3, the transmission parameters in the potential interference mode are dynamically adjusted based on the adaptive weight transmission optimization function, wherein the transmission parameters include baud rate, data bit length, check mode and retransmission strategy.

[0057] In the embodiment, as described in step S1, the electromagnetic interference sources in the industrial field have diversity and complexity, such as pulse interference generated by motor start-stop, high-frequency noise caused by radio frequency equipment, etc. Single-dimensional monitoring is difficult to fully reflect the interference condition. Therefore, in this step, the multi-dimensional interference feature acquisition module integrated in the wireless module is used to synchronously implement real-time acquisition of four key data: the RSSI value is obtained by using the signal strength detection unit to represent the background noise level of the electromagnetic environment; the number of error bits in a unit time is calculated by using the error rate calculation module; the phase difference between the local clock and the reference clock is monitored by using the phase-locked loop (PLL) circuit to identify the low-frequency clock synchronization interference; the electromagnetic radiation in the frequency band of 10 kHz to 100 MHz is analyzed by using the FFT transform algorithm to distinguish the high-frequency transient interference (such as switch power supply pulse) and the medium-frequency periodic interference (such as frequency converter harmonic). After the above multi-dimensional data is normalized, a multi-scale interference feature vector with a length of 4 is formed according to the dimensions of [RSSI, error rate, clock offset, spectral feature], and is continuously updated at a period of 1 ms to provide dynamic data support for the subsequent interference mode prediction.

[0058] As described in step S2 above, to address the gradient vanishing problem in traditional RNN models when processing long sequences and the high computational complexity of the Transformer architecture, this step employs the Mamba state-space model to achieve linear complexity interference time-series modeling. The processing flow of this model is as follows: First, the multi-scale interference feature vector is preprocessed by a depthwise separable convolution module. After extracting local features through 3×3 depthwise convolution, feature fusion is completed by 1×1 pointwise convolution, reducing the number of parameters by 75% while preserving interference feature details. In the core processing stage, based on the recursive relationship S_t = AS_{t-1} + Bu_t of the state-space model (SSM), the interference sequence of the past 100 cycles and the current feature input state vector are modeled using learnable parameters A and B to achieve long-distance dependency modeling. Simultaneously, a gating mechanism is used to dynamically adjust the weights of features in each dimension (e.g., increasing the weight ratio of spectral features during strong electromagnetic interference). To improve prediction robustness, CRC-16-CCITT verification results are introduced as residual input, and the residual connection mechanism F(x) = H(x) + x is fused bit by bit with the model prediction results to correct prediction bias caused by sudden interference. Finally, after processing by the RMSNorm normalization layer and SiLU activation function, the interference mode prediction for the next 5 to 10 communication cycles is output, covering key parameters such as interference intensity (dB), type (pulse / continuous) and impact range (data bits / clock synchronization).

[0059] As described in step S3 above, in order to solve the problem that traditional fixed parameter strategies cannot adapt to dynamic interference environments, this step constructs a three-level weight dynamic adjustment system: the reliability weight α increases with the increase of interference intensity to strengthen the bit error rate constraint; the efficiency weight β automatically increases when the interference weakens to balance the transmission rate and power consumption; and the compatibility weight γ ensures that the parameter adjustment complies with industrial protocol specifications such as Modbus and RS-485. The specific parameter configuration strategy is as follows: For baud rate, when the interference intensity exceeds 80dB, it is reduced to 2400bps to ensure signal integrity; in the 40-80dB range, it is dynamically adjusted to 57600-115200bps; below 40dB, it is increased to 115200bps. Regarding data bits and verification methods, 7 data bits are used with 16-bit CRC-16-CCITT verification during strong interference; during medium interference, it switches to 8 data bits with even parity; and during weak interference, verification is canceled. The retransmission strategy switches according to the interference level: 5 retransmissions are fixed during weak interference; during medium interference, an exponential backoff algorithm with an initial backoff time of 1ms is enabled; and during strong interference, the interrupt recovery mechanism is prioritized to ensure the transmission of critical instructions. At the hardware coordination level, the interference feature extraction IP core achieves microsecond-level response in the FPGA. The MCU dynamically adjusts the main frequency according to the interference level (200MHz for strong interference / 48MHz for weak interference) and writes the parameters to the serial port controller register via the SPI interface to complete the real-time configuration of the transmission parameters.

[0060] In the present application, without increasing the cost of hardware, the cost of anti-interference serial communication is reduced; at the same time, based on the adaptive weight transmission optimization function, the transmission parameters under the potential interference mode are dynamically weighted and adjusted, and the dynamic change of real-time tracking interference is realized to realize anti-interference communication.

[0061] In an embodiment, the interference data includes interference signal strength, bit error rate, clock offset and electromagnetic radiation spectrum data.

[0062] In this embodiment, the interference signal strength refers to the power strength of the received signal, reflecting the background noise level of electromagnetic interference in the current communication link. Directly reflects the superimposed influence of external electromagnetic interference on the communication signal, such as when the motor starts and stops, the RSSI value will fluctuate significantly when the radio frequency equipment is working.

[0063] The bit error rate is the proportion of the number of transmission errors per unit time to the total number of transmission bits, which is the core index to measure the communication quality. Real-time reflects the actual influence of interference on data transmission, such as when electromagnetic interference causes signal distortion, BER will rise sharply; provides a direct basis for transmission parameter adjustment, for example, when BER exceeds 0.001%, trigger baud rate reduction or check mechanism enhancement.

[0064] The clock offset is the phase difference or frequency deviation between the local clock of the wireless module and the standard reference clock, usually in ppm (parts per million) or milliseconds (ms). Identify low-frequency clock synchronization interference, such as power supply ripple, electromagnetic interference near the crystal oscillator will cause clock offset; affect the timing accuracy of serial communication (such as start bit / stop bit recognition of UART), large offset will cause data frame parsing error.

[0065] The electromagnetic radiation spectrum data is obtained by frequency domain analysis of electromagnetic radiation signals through fast Fourier transform (FFT), and the energy distribution of different frequency points is obtained. Distinguish interference types (such as high-frequency transient interference, medium-frequency periodic interference), for example, the pulse interference generated by the switching power supply is concentrated in the 100kHz-10MHz frequency band, while the harmonic interference of the frequency converter is mainly distributed in the 10kHz-100kHz; locate the characteristics of the interference source, provide frequency domain dimension timing modeling materials for Mamba model, improve the prediction accuracy of complex interference mode

[0066] In an embodiment, the multi-dimensional interference feature acquisition module adopts a parallel multi-resolution feature processing structure, including:

[0067] High-frequency interference feature branch, used for capturing transient electromagnetic interference signals above 100kHz;

[0068] Medium-frequency communication quality branch, used for analyzing the bit error rate fluctuation law of 10kHz-100kHz;

[0069] Low-frequency clock offset branch for monitoring clock synchronization deviation below 10 kHz;

[0070] Each branch realizes information fusion through a cross-scale feature interaction module to obtain a multi-scale interference feature vector.

[0071] In this embodiment, the high-frequency interference feature branch is used to capture transient electromagnetic interference above 100 kHz. High-frequency transient interference in industrial environments mainly comes from switching power supplies, radio frequency equipment, etc., and has the characteristics of steep rising edge, short duration (microsecond level), and energy concentrated in the frequency band above 100 kHz. The following processing flow is used in this branch:

[0072] Pre-bandpass filtering: a bandpass filter composed of an LC resonant circuit is used to limit the input signal bandwidth to the 100 kHz-10 MHz interval and suppress low-frequency noise interference;

[0073] High-speed sampling circuit: an ADC (such as ADS8320, sampling rate 20 MSPS) with a sampling rate above 10 MHz is used to digitize the filtered signal, ensuring that nanosecond-level pulse interference is captured;

[0074] Feature extraction algorithm: perform short-time Fourier transform (STFT) on the sampled data to calculate the frequency spectrum energy distribution in each 1 ms time window and extract feature parameters such as peak frequency, pulse width, and rising edge slope;

[0075] Quantitative output: normalize the extracted feature parameters to form a high-frequency interference feature vector with dimensions [peak frequency, pulse energy, pulse density].

[0076] The medium-frequency communication quality branch is used to analyze the 10 kHz-100 kHz bit error rate fluctuation. Interference in the 10 kHz-100 kHz frequency band mainly affects the modulation and demodulation process of communication signals, resulting in bit error rate fluctuation. The branch realizes accurate analysis through the following steps:

[0077] Adaptive bandwidth filtering: use a digital programmable filter (such as MAX262) to dynamically adjust the passband range, focusing on tracking interference signals in the 10 kHz-100 kHz interval;

[0078] Real-time bit error rate calculation: use a sliding window mechanism (window size 1024 bytes) at the receiving end to compare the received CRC check code with the local calculation result, and update the bit error rate value every 10 ms;

[0079] Fluctuation feature extraction: perform wavelet transform on the continuous 100 bit error rate samples to decompose the fluctuation components of different scales and identify periodic fluctuations (such as interference caused by motor rotation) and sudden fluctuations (such as relay attraction);

[0080] Feature vector generation: Combine parameters such as fluctuation frequency, amplitude, duration, etc. into a medium frequency quality feature vector of [fluctuation main frequency, maximum amplitude, fluctuation period].

[0081] Low-frequency clock offset branch is used to monitor the clock synchronization deviation below 10 kHz. Low-frequency interference below 10 kHz mainly affects the stability of the crystal oscillator, causing clock synchronization deviation. This branch uses phase-locked loop (PLL) and digital frequency synthesis (DDS) technology to achieve high-precision monitoring:

[0082] Dual-PLL architecture: The main PLL (such as Si5351) locks the reference clock, and the auxiliary PLL tracks the local crystal oscillator output, and compares the phase difference of the two clock signals in real time through the phase detector (PD);

[0083] Phase error digitization: Use 24-bit sigma-delta ADC (such as ADS1255) to sample the phase error, with a resolution of 0.1° phase deviation;

[0084] Frequency drift calculation: Process the continuously sampled phase error through Kalman filter to separate the crystal oscillator frequency drift (unit: ppm) and time domain jitter (unit: ps);

[0085] Feature quantization output: Generate a low-frequency clock feature vector containing [frequency drift rate, maximum jitter value, drift trend].

[0086] The feature vectors extracted by each branch need to be complementary through cross-scale interaction. This module uses the following mechanisms:

[0087] Feature scaling alignment: Align the high-frequency (1ms sampling period), medium-frequency (10ms), and low-frequency (100ms) feature vectors to the 1ms time scale through linear interpolation;

[0088] Attention mechanism fusion: Use a multi-head attention structure to calculate the correlation coefficient matrix of each branch feature and automatically assign weights.

[0089] Residual fusion path: Preserve the original feature vector while superimposing the attention-weighted fusion feature to form a four-dimensional multi-scale interference feature vector of [high-frequency feature, medium-frequency feature, low-frequency feature, fusion feature];

[0090] Dynamic update mechanism: Adjust the fusion weight adaptively according to the interference intensity, such as increasing the weight of the high-frequency feature branch when there is strong high-frequency interference, to ensure optimal feature expression ability in different interference scenarios.

[0091] Through this parallel multi-resolution structure, the module can simultaneously capture the time / frequency domain characteristics of different frequency band interferences, and achieve information complementation through cross-scale interaction, ultimately outputting a multi-scale feature vector that comprehensively represents the characteristics of industrial interference, providing high-quality input for subsequent time series prediction of the Mamba model.

[0092] In an embodiment, the Mamba communication state space model is specifically used for:

[0093] A deep separable convolution module is used to extract local features from the input multi-scale interference feature vector, obtaining a reduced dimension interference feature.

[0094] A dynamic weight calculation unit of the Mamba state space module is used to perform long-distance dependence modeling on the preset historical interference sequence and the reduced dimension interference feature, and a linear complexity time series convolution operation is used to predict the interference pattern in the next 5 to 10 communication periods.

[0095] The check result of the traditional CRC check code is input as a residual error, and is bitwise fused with the predicted interference pattern, and the predicted deviation is corrected through a residual connection mechanism to obtain a potential interference pattern; the generating polynomial of the CRC check is based on a preset industrial standard.

[0096] In this embodiment, first, the interference characteristics in an industrial environment have high dimensionality and strong correlation, and direct input will result in excessive model calculation burden. This step uses deep separable convolution for feature optimization:

[0097] Feature grouping processing: the input multi-scale interference feature vector is divided into three subspaces of high frequency (above 100 kHz), medium frequency (10-100 kHz), and low frequency (below 10 kHz) according to frequency bands, and each subspace contains time domain, frequency domain, intensity, etc. dimensions;

[0098] Deep convolution operation: a 3x3 deep convolution kernel is applied to each subspace, and independent convolution is performed along the feature dimension direction to extract local dependencies within each frequency band, and the parameter quantity is reduced by 89% compared with standard convolution;

[0099] Point-by-point convolution fusion: a 1x1 point-by-point convolution is used to fuse the feature maps output by the deep convolution across channels, generating a reduced dimension interference feature with a dimension compressed to 60% of the original, while retaining key interference pattern features;

[0100] Batch normalization: a BatchNorm layer is applied to normalize the reduced dimension feature, stabilizing the model training process and accelerating the convergence speed.

[0101] The traditional RNN model has a gradient vanishing problem when processing long sequences, and the self-attention mechanism of the Transformer has a computational complexity of O(n²). In this embodiment, the linear complexity time convolution mechanism of the Mamba model is used:

[0102] State space representation: The historical interference sequence (dimensionality reduction features of the past 100 communication periods) and the current input feature are represented as a state vector S_t, and the time evolution is realized through the state update equation S_t = AS_{t-1} + Bu_t, where A is the state transition matrix and B is the input transformation matrix;

[0103] Dynamic weight calculation: The weights of the features at each historical time are adaptively adjusted through the gating mechanism;

[0104] Linear complexity optimization: Shift operation is used instead of matrix multiplication to reduce the time convolution calculation complexity from O(n³) to O(n), which is realized through the frequency domain fast algorithm of polynomial multiplication;

[0105] Interference pattern prediction: Based on the state vector at the current time t, the interference intensity, type and influence range prediction vector for the next 5-10 communication periods is generated through a fully connected layer.

[0106] Furthermore, the CRC residual fusion mechanism is used to correct the prediction bias. Traditional deep learning models are prone to prediction bias when facing sudden interference. In this step, the CRC check code is introduced as a physical layer verification signal:

[0107] CRC check generation: Calculate the check code C_t for the current transmission data frame according to the preset industry standard (such as CRC-16-CCITT for Modbus protocol);

[0108] Prediction residual calculation: Map the predicted interference pattern P_t to the theoretical check code C'_t, and calculate the residual ΔC = C_t - C'_t;

[0109] Residual correction prediction: ΔC is integrated into the state vector update at the next time through the residual connection mechanism;

[0110] Bit-by-bit fusion strategy: Perform XOR operation between the residual information and the prediction vector bit by bit, focusing on correcting the prediction bias in the high confidence area.

[0111] In this embodiment, the depth separable convolution module is deployed in the DSP slice of the FPGA to realize microsecond-level feature extraction; the Mamba state space module runs on the Cortex-M7 kernel of the MCU to accelerate matrix operations using SIMD instructions; to address the limited computing resources of industrial-grade MCUs, the gradient accumulation technique is used to accumulate the gradients of multiple small batches before updating the parameters, which is equivalent to expanding the batch size; 16-bit fixed-point operations are used during the interference flat period to reduce power consumption, and 32-bit floating-point operations are switched to during the strong interference period to ensure accuracy, and the fast switching is realized through the hardware floating-point unit (FPU). Through the above mechanism, the Mamba communication state space model effectively captures the long-range dependence relationship of industrial interference while maintaining linear computational complexity, and significantly improves the prediction accuracy of sudden interference by combining the CRC residual correction mechanism, providing a reliable basis for subsequent dynamic adjustment of transmission parameters.

[0112] In an embodiment, the Mamba communication state space model further enhances the non-linear expression capability through the RMSNorm normalization layer and the SiLU activation function, and the overall parameter quantity is controlled within 1MB, which is suitable for the embedded computing resources of industrial-grade wireless modules.

[0113] In this embodiment, the memory resources of industrial-grade wireless modules are limited, and the traditional batch normalization layer occupies too much space because it needs to maintain global statistics. The model uses the RMSNorm normalization technique for optimization. This technique only calculates the root mean square value of the input data, not the mean and variance, reducing the computational load by half; the learnable scaling parameter is initialized to 1, which is automatically optimized through training to avoid introducing additional bias parameters; when implemented in FPGA hardware, the lookup table is used to precompute the square root function, further improving the computational efficiency. Compared with traditional batch normalization, RMSNorm can reduce about 60% of memory usage, and the execution speed on industrial-grade MCUs such as STM32H7 is improved by 40%.

[0114] To solve the problem of gradient vanishing in the negative value region of the traditional ReLU activation function, while avoiding the high computational cost of complex functions, the model uses the SiLU activation function. This function is implemented on the MCU through piecewise linear approximation: when the input is less than -6, the output is 0, and when the input is greater than 6, the output is 1, and the middle region is calculated by the linear function 0.125x+0.5. With the SIMD instructions of the MCU, multiple groups of data can be processed in parallel, effectively improving the computational efficiency. The SiLU function has both smoothness and non-linear expression ability, and it retains non-zero gradients in the negative value region, making the model more sensitive to weak interference signals. Experiments show that it can improve the interference prediction accuracy by about 15% compared to ReLU.

[0115] To adapt to the limited resources of industrial-grade MCUs (typical Flash space less than 4MB, RAM less than 1MB), the model is optimized in multiple ways: deep separable convolution is used instead of standard convolution, reducing parameter quantity by more than 75%; L1 regularization and channel pruning algorithm are used in the training stage to delete low-contribution connections and channels, compressing the number of channels to 50% of the original; 8-bit integer quantization technology is introduced, with 16-bit precision reserved for key layers, and a mixed-precision inference engine is developed; the state dimension of the Mamba state space module is compressed to 64 dimensions, the number of attention heads is reduced to 4, and the sequence length is limited. After optimization, the total parameter quantity of the model is compressed from 1.28MB to 320KB, with a compression rate of 75%.

[0116] In terms of memory management, model parameters are stored in Flash, loaded into RAM on demand during runtime, and intermediate variables are managed using memory pools to avoid fragmentation; through graph fusion technology, RMSNorm is combined with the linear transformation of the previous layer and the SiLU activation function with the convolution operation into a single operation, reducing memory access times; at the hardware level, a deep separable convolution dedicated computing unit is deployed on the FPGA, and the DMA controller of the MCU is used to realize zero-copy data transmission. These optimizations enable the model to have an inference delay of less than 5ms and a power consumption of less than 10mW on an STM32H7 MCU, meeting the needs of industrial-grade modules for stable operation 7x24 hours.

[0117] In an embodiment, the adaptive weight transmission optimization function expression is:

[0118]

[0119] wherein, is the bit error rate loss term, using Smooth L1 loss to measure the deviation of the actual bit error rate from the threshold value;

[0120] is the serial port resource consumption loss term, used to constrain the power consumption fluctuations caused by baud rate adjustment;

[0121] is the protocol compatibility loss term, ensuring that parameter adjustment conforms to the industrial communication protocol specification;

[0122] is the dynamic weight coefficient, automatically adjusting the proportion according to the real-time interference intensity.

[0123] In an embodiment, the adjustment strategy of the dynamic weight coefficient includes:

[0124] When the electromagnetic interference intensity exceeds 80dB, set ;

[0125] When the electromagnetic interference intensity is less than 40dB, set ;

[0126] When the electromagnetic interference intensity is not less than 40dB and not more than 80dB, set

[0127] Wherein, the adjustment period of the weight coefficient is inversely proportional to the baud rate of the serial communication.

[0128] In an embodiment, based on the adaptive weight transmission optimization function, the transmission parameters under potential interference mode are dynamically weighted and adjusted, including:

[0129] Obtain the electromagnetic interference intensity, and set the proportions of reliability weight α, efficiency weight β and compatibility weight γ based on the mapping rule of the interference intensity and the weight coefficient; wherein α corresponds to the error rate constraint intensity, β corresponds to the adjustment priority of the baud rate and the data bit length, and γ corresponds to the industrial protocol compatibility verification weight;

[0130] Determine the differentiated adjustment strategy for different transmission parameters:

[0131] Baud rate adjustment: when the α weight proportion is greater than or equal to 3, start the stepwise speed reduction mechanism, and adjust according to 90%, 80% and 70% of the current baud rate, and the interval of each adjustment is not more than 20ms;

[0132] Data bit length configuration: when β is greater than 4, use 8-bit data bit to improve transmission efficiency, otherwise use 7-bit data bit to reduce error risk;

[0133] Verification mode switching: when γ is greater than or equal to 3, CRC-16 verification is forced to be enabled and two parity check bits are added;

[0134] Re-transmission strategy optimization: based on the weight difference between α and β, dynamically adjust the retransmission interval, if the difference is greater than 1, shorten the retransmission interval from 50ms to 30ms, and if the difference is less than or equal to 1, keep the default interval;

[0135] Real-time effect of parameter adjustment is realized through hardware register mapping table, and the weight calculation result is converted into configuration instruction of serial controller.

[0136] In this embodiment, the electromagnetic interference intensity in the industrial environment changes in real time, first through the RSSI (received signal strength indication) sensor to obtain the current interference intensity value, and then map it to the preset weight coefficient space.

[0137] The specific rules are: when the interference intensity is lower than 40dB, the reliability weight α: efficiency weight β: compatibility weight γ = 2:5:3 is set, the transmission efficiency is preferentially guaranteed; when the interference intensity is in the interval of 40-80dB, it is adjusted to 3:4:3, the reliability and efficiency are balanced; when the interference intensity exceeds 80dB, the proportion of 5:3:2 is adopted, the transmission accuracy is mainly ensured. This dynamic weight distribution mechanism can automatically adjust and optimize the target according to the environmental changes.

[0138] When the reliability weight α accounts for 3 or more, the baud rate adaptive adjustment mechanism is started. The mechanism adopts a step-down strategy, adjusts according to 90%, 80%, and 70% of the current baud rate, and strictly controls the interval of each step within 20ms to ensure a quick response to sudden interference. For example, when strong interference is detected, it will quickly decrease from 115200bps to 103680bps, and if the interference continues to exist, it will continue to decrease to 92160bps, and finally to 80640bps. This gradual adjustment method can effectively resist interference and avoid the loss of transmission efficiency caused by excessive speed reduction.

[0139] The efficiency weight β directly affects the selection of data bit length. When β>4, it is determined that the current environment has weak interference, and 8-bit data bit configuration is adopted to maximize transmission efficiency; otherwise, when β≤4, it is switched to 7-bit data bit mode, which reduces the amount of data per frame to reduce the error code risk. This dynamic adjustment mechanism balances transmission efficiency according to environmental conditions while ensuring transmission reliability, and is particularly suitable for industrial scenarios with large fluctuations in interference intensity.

[0140] The compatibility weight γ determines the strictness of the verification method. When γ≥3, CRC-16 verification is forced to be enabled and two parity check bits are added to form double verification protection. CRC-16 verification uses an industry standard generator polynomial, which can detect 99.998% of random errors, and the additional parity check bits further enhance the detection capability for single-bit errors. This multiple verification mechanism significantly improves the reliability of data transmission in a strong interference environment, ensuring that the compatibility requirements of the industrial protocol are met.

[0141] The difference between the reliability weight α and the efficiency weight β is calculated to adjust the retransmission interval. When the difference is >1, it indicates that the current environment needs more aggressive reliability protection, and the retransmission interval is shortened from the default 50ms to 30ms, increasing the number of retransmissions per unit time and improving the data recovery probability; when the difference is ≤1, the default retransmission interval is maintained to balance power consumption and efficiency. This dynamic adjustment strategy based on the weight difference allows the system to intelligently allocate resources under different interference intensities, avoiding unnecessary energy consumption.

[0142] To ensure real-time effectiveness of parameter adjustments, a hardware register mapping table was established to directly convert software-calculated weight parameters into configuration instructions for the serial port controller. Specifically, configuration information such as baud rate, data bit length, and parity mode are written to the STM32 microcontroller's USART_CR1 / CR2 registers via the SPI interface, while retransmission strategy parameters are written to the DMA controller's configuration register. The entire parameter update process is completed within 2ms, ensuring rapid response to changes in interference. This hardware-level real-time configuration mechanism achieves industrial-grade real-time requirements while maintaining software flexibility.

[0143] This adaptive weighted transmission optimization mechanism achieves an intelligent balance between transmission reliability and efficiency in industrial environments with dynamically changing interference intensity through the coordinated adjustment of multi-dimensional parameters. Actual measurements show that the bit error rate is reduced by 35% under strong interference conditions, while maintaining a transmission efficiency advantage of more than 15%.

[0144] In one embodiment, after dynamically weighting and adjusting the transmission parameters under potential interference modes based on an adaptive weighted transmission optimization function, the process includes:

[0145] Obtain the identification number of the industrial-grade wireless module, and generate an identification character array based on the identification number;

[0146] Obtain a first preset encoding table and superimpose it with the identifier character array according to rules. Based on the positional relationship between the array elements and the encoding table elements, determine multiple first numeric characters and multiple non-numeric characters; generate a first data carrier based on each first numeric character.

[0147] Obtain a second preset encoding table, and mutate the second preset encoding table based on the first data carrier to obtain a mutated encoding table; decode the non-numeric characters based on the mutated encoding table to obtain a plurality of corresponding second numeric characters;

[0148] A second data carrier is generated based on each second numeric character, a third preset encoding table is obtained, and multiple third numeric characters are generated based on the second data carrier and the third preset encoding table.

[0149] The first, second, and third digit characters are combined to obtain a number combination, which serves as the communication key for the industrial-grade wireless module.

[0150] In this embodiment, firstly, the identification number of the industrial-grade wireless module is obtained. This identification number is a unique identifier for the module in the network, similar to a device's "ID number." Based on this identification number, the system converts it into an array of identifying characters using a specific algorithm. This array contains characterized information about the module's identity, providing the basic data for subsequent key generation.

[0151] Then, a first preset encoding table is called, which is a set of pre-set encoding rules. The identification character array is superimposed with the first preset encoding table according to a predetermined rule, and a plurality of first digital characters and a plurality of non-digital characters that meet the conditions are screened out by analyzing the positional relationship between the array elements and the encoding table elements. These screened characters contain key information after the interaction of the module identification and the encoding rules, and then a first data carrier (graph, curve, matrix, data table, etc.) is generated based on each first digital character, which carries the effective data after preliminary processing and lays the foundation for subsequent steps.

[0152] Then, a second preset encoding table is obtained, which is also a pre-set encoding rule, but has a different function from the first preset encoding table. Based on the first data carrier generated before, the second preset encoding table is subjected to mutation operation, and the structure or parameters of the encoding table are changed through the information in the data carrier to obtain a mutated encoding table. After the formation of the mutated encoding table, the non-digital characters screened out before are decoded using the mutated encoding table to convert the non-digital characters into a plurality of second digital characters, further mining and refining the key information.

[0153] After that, a second data carrier (graph, curve, matrix, data table, etc.) is generated based on each second digital character, which further integrates the newly generated effective data. At this time, a third preset encoding table is obtained, and the second data carrier and the third preset encoding table are combined to generate a plurality of third digital characters, and the data is processed and converted again.

[0154] Finally, each first digital character, second digital character and third digital character obtained before is combined, and these digital characters are fused into a complete digital combination through a specific permutation and combination method. This digital combination is finally used as the communication key of the industrial-grade wireless module for data encryption, decryption and identity verification operations in subsequent communication processes to ensure communication security and reliability.

[0155] In an embodiment, after dynamically weighting and adjusting the transmission parameters under potential interference mode based on the adaptive weight transmission optimization function, it includes:

[0156] Obtain the real-time environment data of the industrial-grade wireless module, normalize it and construct an environment parameter vector;

[0157] Call a preset basic encoding template, superimpose the environment parameter vector and the basic encoding template in a spiral manner, and screen out a plurality of basic digital characters and control symbols according to the phase difference relationship between the elements; use the basic digital characters to generate an initial key segment;

[0158] Obtain the working mode characteristics of the industrial-grade wireless module, perform topological deformation on the basic encoding template to generate a dynamic encoding network; parse the control symbols through the dynamic encoding network to convert them into multiple extended digital characters;

[0159] Combine the initial key fragment with the extended digital characters to form an intermediate key, and perform cyclic shift and modulo operation on the intermediate key based on the operating cycle characteristics of the industrial field device to generate a final digital sequence;

[0160] Extract the hardware characteristics of the industrial-grade wireless module, quantize them into feature codes, and hash fuse them with the final digital sequence to output as the communication key of the industrial-grade wireless module.

[0161] In this embodiment, first, various data in the environment where the industrial-grade wireless module is located are collected in real time, including but not limited to electromagnetic interference intensity, environmental temperature, humidity, air pressure, etc. These original data units and magnitudes are different, in order to facilitate subsequent processing, normalization operation is performed on them, all data are mapped to a specific interval to eliminate the dimension influence. Subsequently, based on the normalized data, an environmental parameter vector is constructed, which integrates the environmental information in a structured form and provides basic data support for key generation.

[0162] Then, the pre-set basic encoding template is called, which contains fixed encoding rules and data mapping relationship. The environmental parameter vector and the basic encoding template are spirally superimposed, which is not a simple data merging, but an interweaving of elements according to the spiral track. In the superposition process, the system selects multiple basic digital characters and control symbols that meet the specific conditions according to the phase difference relationship between the elements. Among them, the basic digital characters contain key information of environmental data, and the initial key fragment is generated using these characters, which is the basic component of the key and preliminarily lays the framework of the key.

[0163] Then, the current working mode characteristics of the industrial-grade wireless module are obtained, such as data transmission rate, communication protocol type, device operating state, etc. Based on these working mode characteristics, the system performs topological deformation on the basic encoding template to change the structure and connection relationship of the template, thereby generating a dynamic encoding network. This network can adapt to the needs of the module under different working modes. The control symbols selected previously are parsed through the dynamic encoding network, and the control symbols are converted into multiple extended digital characters according to the connection and conversion rules in the network, further enriching the information content of the key and increasing the complexity and security of the key.

[0164] Then, the initial key fragment is combined with the extended digital character to form an intermediate key. At this time, the intermediate key has fused the environment and work mode related information, but in order to make it more in line with the actual needs of the industrial site, the running cycle characteristics of the industrial site equipment such as device start-stop time, operation cycle length, etc. are combined to perform cyclic shift and modulo operation processing on the intermediate key. Cyclic shift changes the arrangement order of the digital character, and modulo operation limits and converts the numerical value. After this series of operations, the final digital sequence is generated, which is further improved in security and applicability.

[0165] Finally, the hardware features of the industrial-grade wireless module are extracted, such as chip model parameters, antenna frequency response characteristics, circuit board wiring features, etc., and these hardware features are quantified into feature codes to represent the hardware uniqueness of the module in digital form. The feature code and the final digital sequence are fused by a hash function, and the one-way and confusion characteristics of the hash function are used to mix the information of the two to generate a fixed-length output value. This output value is the communication key of the industrial-grade wireless module. This key combines environment, work mode, device running cycle and hardware features, etc. multi-dimensional information, has strong uniqueness and security, and can effectively protect the communication security of the module in the industrial environment.

[0166] In an embodiment, based on the adaptive weight transmission optimization function, the transmission parameters under potential interference mode are dynamically weighted and adjusted, including:

[0167] Obtain the power supply voltage fluctuation curve, crystal oscillator frequency offset historical data and wireless channel fading depth distribution of the industrial-grade wireless module, and construct the module running state feature map;

[0168] Call the fourth preset encoding table, topologically associate the key parameter nodes in the module running state feature map with the elements of the fourth preset encoding table, and select a plurality of basic key numbers and check identifiers through connection weight calculation between nodes; generate a key core segment using the basic key numbers;

[0169] Monitor the temperature and humidity gradient change, dust concentration fluctuation and equipment vibration acceleration peak value of the industrial site, and use the environment change prediction model to predict the environment change, and according to the prediction result, the characteristic value of the fourth preset encoding table is scaled and the elements are replaced to generate a dynamic encryption table; use the dynamic encryption table to perform transcoding processing on the check identifier to obtain a key extension segment;

[0170] Modularly splice the key core segment and the key extension segment to form a temporary key architecture, and then rearrange the temporary key architecture according to the stroboscopic frequency of the on-site lighting equipment;

[0171] The antenna pattern offset of the wireless module, the code interference intensity of the received signal are collected, which are quantized as a signal feature sequence, and a reorganized key architecture is subjected to bit operation encryption, and finally the communication key of the industrial wireless module is generated.

[0172] In the embodiment, first, the power supply voltage fluctuation curve (reflecting power supply stability), the crystal oscillator frequency offset history data (affecting clock synchronization accuracy), and the fading depth distribution of the wireless channel (reflecting signal transmission quality) of the industrial wireless module are collected in real time. The three types of data are subjected to spatiotemporal alignment and normalization processing, and the voltage fluctuation amplitude, frequency offset, and channel fading depth are used as three-dimensional coordinate axes to construct a module operating state feature map containing time domain dynamic characteristics. The map presents the comprehensive operating state of the module in the electromagnetic interference environment through topological structure visualization, providing bottom physical feature support for key generation.

[0173] Then, the fourth preset encoding table (a topological encoding rule set based on graph theory) is called to establish a weighted association between the key parameter nodes (such as voltage mutation points, crystal oscillator offset extreme points, and channel fading critical points) in the feature map and the encoding table elements. By calculating the connection weight between the nodes (the weight value is determined by the parameter influence degree and the interference correlation), the node mapping results with a value greater than the threshold value are screened out to generate a basic key number and a check identifier. The basic key number forms a key core segment after binary encoding, which carries the core feature information of the module operating state and constitutes the main framework of the communication key.

[0174] Further, the temperature and humidity gradient changes, dust concentration fluctuations, and equipment vibration acceleration peak values in the industrial field are monitored in real time, and the data are input into an environmental change prediction model (based on LSTM neural network training) to predict the environmental parameter trends in the next 50 ms. According to the prediction results, the fourth preset encoding table is subjected to double transformation:

[0175] Feature value scaling: adjusting the numerical interval of the encoding table elements according to the temperature and humidity change rate;

[0176] Element permutation: using the dust concentration peak value as a seed parameter, rearranging the row and column order of the encoding table through the Fisher-Yates shuffle algorithm to generate a dynamic encryption table.

[0177] Subsequently, the dynamic encryption table is used to perform path transcoding on the check identifier (mapping the identifier characters to a multi-dimensional vector and then performing matrix operation), obtaining a key extension segment containing environmental prediction information, which forms a time-space dimension information complement with the key core segment.

[0178] Then, the key core segment and the extension segment are spliced as a temporary key architecture according to functional modules (the first 16 bits are a running state code, and the last 16 bits are an environment prediction code). Since the strobe frequency (such as 50Hz fluorescent lamp) of the industrial field lighting device can cause periodic interference to the wireless signal, the system divides the temporary key architecture into equal-length data blocks according to the strobe phase difference, and randomly reorganizes the data blocks (such as moving the third block to the first position in the nth period), so that the key structure is dynamically aligned with the field interference period, and the anti-interference ability is enhanced.

[0179] Finally, the antenna directivity diagram offset (radiation mode change affected by temperature) of the wireless module and the code interference intensity of the received signal (reflecting the channel distortion degree) are collected and quantized as an 8-bit binary signal feature sequence. The sequence is encrypted bit by bit with the reorganized key architecture through XOR operation, and then a 32-bit final key is generated after SHA-256 hash compression. In this process, the hardware features are used as the input of the physically unclonable function (PUF), which ensures that the key is uniquely bound to the module hardware, and the bit operation encryption integrates the real-time signal distortion features, so that the key has a threefold security mechanism of "environment perception-hardware binding-dynamic encryption".

[0180] Through the above steps, the generated communication key not only integrates dynamic information such as module running state and environment prediction, but also realizes the cooperative optimization of anti-interference ability and security through topological association, dynamic coding table, hardware feature binding and other technologies, which is suitable for industrial-grade wireless communication scenarios in strong electromagnetic interference environment.

[0181] With reference to Figure 2 In another embodiment of the present application, an anti-interference serial communication system of an industrial-grade wireless module is also provided, comprising:

[0182] The collection unit is configured to collect interference data in the serial communication of the industrial-grade wireless module in real time based on the multi-dimensional interference feature collection module, and form a multi-scale interference feature vector.

[0183] The prediction unit is configured to input the multi-scale interference feature vector into the Mamba communication state space model with the fusion CRC check mechanism, and predict a potential interference mode through time sequence modeling of linear complexity.

[0184] The adjustment unit is configured to dynamically weight and adjust the transmission parameters under the potential interference mode based on the adaptive weight transmission optimization function, wherein the transmission parameters include baud rate, data bit length, check mode and retransmission strategy.

[0185] In the present embodiment, the specific implementation of each unit in the above system embodiment is described in the above method embodiment, which will not be described here.

[0186] With reference toFigure 3 The computer device can be a server, and an internal structure of the computer device can be as shown in the figure. Figure 3 The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store corresponding data in the embodiment. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the above method.

[0187] Those skilled in the art can understand that Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied.

[0188] The computer readable storage medium in the embodiment of the present application can be a volatile readable storage medium or a non-volatile readable storage medium.

[0189] In summary, the anti-interference serial communication method and system of the industrial-grade wireless module provided in the embodiment of the present application include: a multi-dimensional interference feature acquisition module, which is configured to collect interference data in serial communication of the industrial-grade wireless module in real time to form a multi-scale interference feature vector; the multi-scale interference feature vector is input into a Mamba communication state space model with a fusion CRC check mechanism, and a potential interference mode is predicted through time sequence modeling of linear complexity; based on an adaptive weight transmission optimization function, transmission parameters in the potential interference mode are dynamically weighted and adjusted; the transmission parameters include baud rate, data bit length, check mode and retransmission strategy. In the present application, the cost of anti-interference serial communication is reduced without increasing hardware cost. At the same time, based on the adaptive weight transmission optimization function, the transmission parameters in the potential interference mode are dynamically weighted and adjusted, so that real-time tracking of dynamic changes of interference is realized to realize anti-interference communication.

[0190] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium provided by the present application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0191] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, device, article or method that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, device, article or method. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, device, article or method that includes the element.

[0192] The above description is only the preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, based on the content of the present application specification and drawings, are also included in the patent protection scope of the present application.

Claims

1. An anti-interference serial communication method of an industrial-grade wireless module, characterized in that, The method comprises the following steps: Based on the multi-dimensional interference feature acquisition module, the interference data in the industrial-grade wireless module serial communication is collected in real time to form a multi-scale interference feature vector; The multi-scale interference feature vector is input into the Mamba communication state space model with CRC check mechanism, and the potential interference mode is predicted through linear complexity time sequence modeling; The specific Mamba communication state space model is used for: adopting a deep separable convolution module to perform local feature extraction on the input multi-scale interference feature vector to obtain a reduced dimension interference feature; Through the dynamic weight calculation unit of the Mamba state space module, the long-distance dependence modeling is performed on the preset historical interference sequence and the reduced dimension interference feature, and the time convolution operation of linear complexity is used to predict the interference mode of the next 5 to 10 communication periods; The check result of the traditional CRC check code is input as a residual error, which is bit-by-bit fused with the predicted interference mode, and the prediction deviation is corrected through the residual connection mechanism to obtain the potential interference mode; wherein the generating polynomial of the CRC check is based on the preset industrial standard; The Mamba communication state space model further enhances the non-linear expression capability through the RMSNorm normalization layer and the SiLU activation function, and the overall parameter quantity is controlled within 1MB, which is suitable for the embedded computing resources of the industrial-grade wireless module; Based on the adaptive weight transmission optimization function, the transmission parameters under the potential interference mode are dynamically weighted and adjusted; wherein the transmission parameters include baud rate, data bit length, check mode and retransmission strategy.

2. The method of claim 1, wherein, The interference data includes interference signal strength, bit error rate, clock offset and electromagnetic radiation spectrum data.

3. The method of claim 1, wherein, The multi-dimensional interference feature acquisition module adopts a parallel multi-resolution feature processing structure, which comprises: A high-frequency interference feature branch for capturing transient electromagnetic interference signals above 100kHz; A medium-frequency communication quality branch for analyzing the bit error rate fluctuation law of 10kHz-100kHz; A low-frequency clock offset branch for monitoring the clock synchronization deviation below 10kHz; Each branch realizes information fusion through a cross-scale feature interaction module to obtain a multi-scale interference feature vector.

4. The method of claim 1, wherein, The adaptive weight transmission optimization function expression is: wherein, is the bit error rate loss term, Smooth L1 loss is used to measure the deviation of the actual bit error rate from the threshold value; serial resource consumption loss term for constraining power consumption fluctuation caused by baud rate adjustment; For the protocol compatibility loss term, ensure that the parameter adjustment conforms to the industrial communication protocol specification; The dynamic weight coefficient is automatically adjusted according to the real-time interference intensity.

5. The method of claim 4, wherein, The adjustment strategy of the dynamic weight coefficient includes: When the electromagnetic interference intensity exceeds 80 dB, set ; When the electromagnetic interference intensity is lower than 40dB, set ; When the electromagnetic interference intensity is not less than 40 dB and not more than 80 dB, set Wherein, the adjustment period of the weight coefficient is inversely proportional to the baud rate of the serial communication.

6. The method of claim 4, wherein, Based on the adaptive weight transmission optimization function, the transmission parameters under the potential interference mode are dynamically weighted and adjusted, including: Obtain the electromagnetic interference intensity, set the proportions of reliability weight α, efficiency weight β and compatibility weight γ based on the mapping rule of interference intensity and weight coefficient; wherein α corresponds to the bit error rate constraint strength, β corresponds to the adjustment priority of baud rate and data bit length, and γ corresponds to the industrial protocol compatibility check weight; Determine the differential adjustment strategy for different transmission parameters: Baud rate adjustment: when the α weight proportion is greater than or equal to 3, start the stepwise speed reduction mechanism, and adjust the baud rate by 90%, 80% and 70% of the current baud rate, and the interval of each adjustment is not more than 20ms; Data bit length configuration: 8-bit data bits are used to improve transmission efficiency when β>4, otherwise 7-bit data bits are used to reduce error risk; Check mode switching: γ≥3 forces to enable CRC-16 check and adds two parity check bits; Re-transmission strategy optimization: based on the weight difference between α and β, the retransmission interval is dynamically adjusted, the difference>1 shortens the retransmission interval from 50ms to 30ms, and the difference≤1 keeps the default interval; Real-time effect of parameter adjustment is realized through hardware register mapping table, and the weight calculation result is converted into serial controller configuration instruction.

7. An anti-interference serial communication system of an industrial-grade wireless module, characterized in that, Comprise: The acquisition unit is used for collecting the interference data in the industrial-grade wireless module serial communication based on the multi-dimensional interference feature acquisition module, and forming a multi-scale interference feature vector; The prediction unit is used for inputting the multi-scale interference feature vector into the Mamba communication state space model of the fusion CRC check mechanism, and predicting the potential interference mode through the linear complexity time sequence modeling; the Mamba communication state space model is specifically used for: adopting a deep separable convolution module to perform local feature extraction on the input multi-scale interference feature vector, and obtaining a reduced dimension interference feature; through a dynamic weight calculation unit of the Mamba state space module, a long distance dependence modeling is performed on the preset historical interference sequence and the reduced dimension interference feature, and a linear complexity time convolution operation is used to predict the interference mode of the next 5 to 10 communication periods; the check result of the traditional CRC check code is input as a residual error, which is bit by bit fused with the predicted interference mode, and the prediction deviation is corrected through the residual connection mechanism to obtain the potential interference mode; wherein the generating polynomial of the CRC check is based on the preset industrial standard; the Mamba communication state space model further enhances the nonlinear expression ability through the RMSNorm normalization layer and the SiLU activation function, and the overall parameter quantity is controlled within 1MB, which is suitable for the embedded computing resources of the industrial-grade wireless module; The adjustment unit is used for dynamically weighting and adjusting the transmission parameters under the potential interference mode based on the adaptive weight transmission optimization function; wherein the transmission parameters include baud rate, data bit length, check mode and retransmission strategy.

Images